Font Size: 

Machine learning (ML) algorithms exhibit powerful performance in many diffenent fields. Nonetheless, designing and training those ML models are still challenging when applied to different types of problems. The collection of training data and hyperparameters tuning play an important role in improving the model accuracy. In this article, a data-collection scheme based on an multi-infill strategy is developed to obtain the maximum information data. Tuning hyperparameters of a ML model are automatically optimized using Bayesian optimization (BO). Accordingly, the obtained ML model is established to replace conventional finite element analyses (FEAs). Differential evolution (DE) algorithm is employed to resolve the structural optimization problems. Several numerical examples are given to demonstrate the efficiency and validity of the proposed method.

Bayesian optimization (BO), Tuning hyperparameter, Neural network, Data-collection, Structural optimization, Machine learning (ML)